The Longitudinal Expert AI Panel (LEAP) released its Wave 8 results this week, and the numbers tell a story of growing consensus among those who track AI's trajectory: things are moving faster than expected. The survey, conducted between April 20 and May 11, 2026, gathered forecasts from 264 participants who completed both the initial Wave 1 (June/August 2025) and this latest round—including AI researchers, superforecasters, policy experts, and members of the general public.

Tracking AI's Impact on the Technological Richter Scale

The survey asks panelists to assess AI's likely impact by 2040 using Nate Silver's Technological Richter Scale (TRS), which ranks technologies from Level 7 ("technology of the decade," like credit cards) up to Level 10 ("technology of the epoch," like the rise of humans). On average, experts now assign a 35% probability to AI reaching TRS Level 8 by 2040—comparable to electricity or the internet—with substantial weight on Level 9 (24%) and even Level 10 (11%). Superforecasters gave a near-identical distribution: 34%, 23%, and 8.5% respectively. The shift from Wave 1 is notable. Among matched participants, the mean expected TRS level rose by +0.20 for experts (from 7.86 to 8.06) but jumped +0.39 for superforecasters (from 7.50 to 7.89). Where "technology of the decade" was the most common modal forecast among superforecasters in Wave 1, "the technology of the century" now dominates at 43%. Experts showed similar movement: those whose modal forecast landed on Level 8 grew from 38% to 53%.

When AGI Arrives—and Who Agrees It Exists

LEAP also pressed panelists on artificial general intelligence. The median expert assigns an 80% probability that more than half of LEAP panelists will agree AGI exists before 2100, defined as a commercially available system that outperforms the 90th percentile human in every primarily non-physical occupation at no more than 5x human labor cost. Conditional on this consensus occurring, experts predict it happening by 2050 at the median (25% probability by 2039, 75% by 2065). Superforecasters are slightly more aggressive: they give the same 80% probability but place the median year at 2047—three years earlier. This divergence between expert and superforecaster timelines has been a consistent pattern across LEAP waves.

METR Benchmark Shows Models Already Outpacing Forecasts

One concrete data point comes from METR's task-completion time horizon benchmark, which measures how long of a human expert task an AI can complete at 80% success. When asked when an AI model will achieve 80% on eight-hour tasks, the median expert predicted 2030 or earlier; superforecasters forecasted 2028, and the public lagged significantly at 2037. On May 8, 2026—toward the end of LEAP's survey period—METR updated its benchmark to include a preview of Anthropic's Mythos model. That model achieved an 80% time horizon of 3 hours and 6 minutes, already within range of expert predictions with more than seven months remaining in 2026. All three groups' median forecasts for end-of-2026 performance landed between 3 and 4 hours, up from a baseline of 1.5 hours at survey launch.

The Optimism Gap—and One Area of Consensus

Panelists diverged sharply on AI's broader societal impact over the next 20 years. When asked about overall effect on the U.S., 57.5% of experts and 69.8% of superforecasters predicted a somewhat or very positive outcome, compared to just 42% of the public. The gap widens further on specific capabilities: 72.7% of experts said AI will make people better at problem-solving versus only 48.6% of the public. On making difficult decisions, 58.3% of experts expect improvement while 39% of the public expects people to become worse at this. The one area where expert and public opinions converge—and not in a good way—is meaningful relationships. Nearly 68% of both experts (68.4%) and the public (66.7%) believe AI will make people worse at forming meaningful connections with others. Superforecasters are slightly more optimistic on this front, with only 50.9% expecting degradation.

Key Takeaways

  • Both experts and superforecasters have meaningfully revised upward their expectations for AI's societal impact by 2040, trending toward "technology of the century" (Level 8) classifications
  • Superforecasters show larger shifts than domain experts, suggesting external validators see faster progress than those inside the field
  • AGI consensus (80% probability) is forecast for 2050 by experts and 2047 by superforecasters—earlier than many public narratives suggest
  • Anthropic's Mythos model already matches 2026 end-of-year benchmark predictions with months remaining in the calendar

The Bottom Line

The LEAP Wave 8 data paints a picture of accelerating progress that even conservative domain experts can no longer dismiss. When superforecasters—who make their living updating beliefs against evidence—show larger upward revisions than people embedded in AI research, that's a signal worth heeding. The real wildcard isn't when transformative AI arrives; it's whether our institutions can adapt to it.